Araucaria: Software for Puzzles in Argument Diagramming

REED & ROWE, ARAUCARIA:ARGUMENT DIAGRAMMING AND XML

1

Araucaria: Software for Puzzles in Argument

Diagramming and XML

CHRIS REED & GLENN ROWE

Department of Applied Computing, University of Dundee, Dundee DD1 4HN Scotland

[chris, growe]@computing.dundee.ac.uk

Abstract. Argumentation theory involves the analysis of naturally occurring argument, and

one key tool employed to this end both in the academic community and in teaching critical

thinking skills to undergraduates is argument diagramming. By identifying the structure of an

argument in terms of its constituents and the relationships between them, it becomes easier to

critically evaluate each part of an argument in turn. The task of analysis and diagramming,

however, is labor intensive and often idiosyncratic, which can make academic exchange

difficult. The Araucaria system provides an interface which supports the diagramming

process, and then saves the result using AML, an open standard, designed in XML, for

describing argument structure. Araucaria aims to be of use not only in pedagogical situations,

but also in support of research activity. As a result, it has been designed from the outset to

handle more advanced argumentation theoretic concepts such as schemes, which capture

stereotypical patterns of reasoning. The software is also designed to be compatible with a

number of applications under development, including dialogic interaction and online corpus

provision. Together, these features, combined with its platform independence and ease of use,

have the potential to make Araucaria a valuable resource to the academic community.

Keywords: Argumentation schemes, argumentation theory, critical thinking, diagramming,

informal logic, XML.

1.Introduction and Motivation


2

Argumentation theory aims to analyse, describe, and evaluate arguments that occur in the real

world, and occurs as a topic in many undergraduate syllabi, where it aims to teach students

both to think critically about the arguments of others, and to create better, more measured

arguments of their own. One of the key tools available to the discipline is diagramming. The

claims and their associated reasons within a given argument are identified, and the

relationships between them drawn up as trees. This diagram then serves as a basis for

criticism and reflection.

Computer software might well be anticipated to be highly suited to the task of visualisation,

and particularly to the sort of diagrams used in argumentation theory. Similarly, it is also

well suited to the task of aiding an analyst in constructing the diagrammatic representation of

an argument. And yet, there are very few computer systems which support argument

diagramming for the student, and none at all which support the diversity and sophistication of

analyses formed within the research community. It is this dearth of computer support for a

labor intensive but crucial activity, which is addressed in this paper.

2.Background

The development of informal logic and argumentation theory within philosophy has

represented a backlash against post-Fregean formal logic, which though immensely powerful

and widely applicable, is a poor choice for representing and characterising natural – i.e. real-

world – language and argument, despite its Aristotelian heritage aimed at just that. The

inception of informal logic marked by Toulmin (1958), Perelman and Olbrechts-Tyteca

(1969), and others saw a return to an empirically driven logic.

Within argumentation theory, systems of diagramming argument have played an important

practical role, in two distinct areas. The first is in pedagogy: employing diagrams in support

of the teaching of critical thinking skills. Though opinion is divided as to the degree to which

diagramming is useful for all students (see, for example, (Argthry, 2001)) there is clear

evidence that the technique is of great benefit for some (van Gelder & Rizzo, 2001). These

results concur with more general results in the psychology of reasoning (see (Rips, 1994: 347-

349) for an introductory discussion of this issue).

The driving force provided by the need to teach critical thinking, and the rise in utility and


3

subsequent popularity of computer assisted learning packages, has led to the appearance of a

number of software systems for argument diagramming which are intended for pedagogical

use. One of the most sophisticated and polished examples is the Reason!Able system (van

Gelder & Rizzo, 2001). This system aids students – particularly those learning informal

reasoning skills at an introductory level in schools and universities (and, they report, even

kindergarten) – in constructing and analysing argument maps. These maps employ arrows and

colours to indicate support and rebuttal relationships and are manipulated through a

straightforward interface that is appealing to children. The preliminary results reported in (van

Gelder & Rizzo, 2001) suggest that the software has, at the very least, the potential to

substantially improve students’ critical thinking skills. For van Gelder, as with the work

described below, the focus is on the software itself; there are other examples where the focus

is on other resources (typically a textbook) and the software plays more of a supporting role.

A good example of this is LeBlanc’s Lemur software which accompanies her text (LeBlanc,

1998), and supports dynamic student interaction with the exercises with which each chapter is

brought to a close. Some of these exercises involve basic diagramming, but as the software is

encompassing the range of techniques introduced in the text, the diagramming components

themselves are rather limited and constraining, and unlike Reason!Able, are restricted to only

those examples provided in the software.

The second role of diagramming is in the construction and implementation of theories of

argument evaluation within the research community. One of the earliest methods, now the

textbook favourite, is that proposed by Beardsley (1950), and enhanced with nomenclature by

Thomas (1986). More recently, inadequacies and problems with this standard treatment have

been identified, leading several authors to propose alternatives, e.g. (Freeman, 1991) which

extends the standard treatment to deal with structures described by Toulmin (1958); (Reed,

1999) which explores an alternative method of handling linked structures; and (Wilson, 1986)

which emphasises the evaluative aspect of argument analysis by including it explicitly in the

diagram. The various approaches to the issue of diagramming, of which these are just a few,

represent scholarly approaches to problems that lie at the heart of argumentation theory and

informal logic.

Although there has not to date been software specifically designed to support research into

argumentation and diagramming, there have been a few systems which impact upon that

research. Foremost amongst these is the ambitious and far-reaching Archelogos project under


4

development at the University of Edinburgh (Scaltsas, 1997), which aims to analyse and

mark-up substantial portions of the argumentation in the oeuvres of both Plato and Aristotle,

and provide an interface that allows online navigation of the structure of the reasoning in the

works. The Archelogos project does not, however, focus upon diagramming the structure that

is produced through analysis. In contrast, work in linguistics, and in particular in pragmatics,

aiming to analyse interclausal relations, uses software tools to build diagrammatic analyses of

textual structure. RSTtool (O’Donnell, 1997) is a good example of such software, and it has

been argued (Mann and Thompson, 1988) that the approach can be applied to argumentative

text (just as it can to any other genre). These linguistic research projects make no use,

however, of the rich analytical structures and techniques of argumentation theory.

Finally, argumentation itself has found many applications within computer science, and

various branches of artificial intelligence in particular. A review of many of these systems can

be found in (Reed, 1997), whilst a more recent analysis of interdisciplinary work between

argumentation and each of multi-agent systems, legal reasoning decision support,

computational linguistics, and contextual reasoning, can be found in (Reed et al., 2002).

The focus here, however, is squarely upon software to support both teaching and research in

argumentation theory. To the authors’ knowledge, there is currently no system which provides

such support, and it is this gap that the Araucaria system fills.

3.Araucaria

The Araucaria system does not attempt to tackle fundamental restrictions of the diagramming

process. As with other methods of analysing textual structure – such as Rhetorical Structure

Theory, RST (Mann and Thompson, 1988), for example – any given analysis is potentially

disputable. RST offers a means of specifying the relation that holds between spans of text –

though both the judgements concerning the delimitation of text structure phrases, and the

identification of relationships between those phrases, can be challenged. Mann and Thompson

suggest that in marking up the rhetorical structure of a text, the analyst makes plausibility

judgements (rather than absolute analytical decisions) and that there can be more than one

reasonable analysis. The assumptions behind Araucaria follow the same pattern: a single text

might be analysed in several different ways, depending upon a variety of analytical choices.


5

As in RST, the judgements concerning the delimitation of argument components can vary,

depending upon the aims of the analyst and the clarity of the text itself. Example (Ex1) is an

excerpt from an extended argument taken from a US Supreme Court case, and used in a

current textbook (Johnson, 1998: 328).

(Ex1) It [the Court's decision] permits the state's abstract,

undifferentiated interest in the preservation of life to

overwhelm the best interests of Nancy Beth Cruzan, interest

which would, according to an undisputed finding, be served

by allowing her guardians to exercise her constitutional right

to discontinue medical treatment. ... Because Nancy Beth

Cruzan did not have the foresight to preserve her

constitutional right in a living will, or some comparable

"clear and convincing" alternative, her right is gone forever

and her fate is in the hands of the state Legislature instead of

in those of her family, her independent neutral guardian ad

litem, and an impartial judge – all of whom agree on the

course of action that is in her best interests.

One defensible analysis would include four components in the first sentence, and three in the

second. Johnson's analysis, in contrast, considers each of the two sentences to be indivisible

units. In the context of the analysis, the reasons for his choice are clear, namely, to ensure that

the analysis is at an appropriate level of abstraction (the text of the example runs over several

pages), and to provide a pedagogically sound tutorial.

Also analogously to RST, decisions about the rela tionships between components may also

vary. Freeman (1991) points out, for example, that a given text might be analysed using the

Toulmin (1958) schema, and the data and warrant be quite interchangeable. Furthermore,

recent work examining argumentation schemes – stereotypical patterns of nondeductive

reasoning – has demonstrated that a single text might be regarded as instantiating several

different schemes, depending on the focus of the analysis (Walton, 1996).

Again by analogy to RST, there is also freedom in analytic resources. Mann and Thompson

emphasise that the set of relations they put forward is simply one possible set that has been


6

found to have utility in the analysis of a particular corpus. They claim neither exhaustiveness

nor accuracy of their proposed set, instead describing the process by which researchers can

produce their own sets of relations. A similar solution is adopted in the provision of

schemesets of argumentation schemes. Many scholars and teachers of critical thinking and

related fields find that argumentation schemes are a useful tool for describing the

relationships between argument components. Determining a single, exhaustive, consistent set

of schemes has proved difficult – though existing sets such as (Grennan, 1996),

(Kienpointner, 1992) and (Walton, 1996) are nevertheless rich and extensive. The importance

of argumentation schemes is also growing within various computational applications of

argument (Reed and Walton, 2001), so one aim in developing the Araucaria software was to

ensure that argumentation schemes were coherently integrated. The choice of which - if any -

argument set to use is left to the user, with the standard distribution including not only

schemeset definitions corresponding to the Grennan, Kienpointner and Walton lists

mentioned, but also software to design custom schemesets in a straightforward manner.

Finally, one analytic freedom with which RST does not have to contend is the process of

reconstruction, and in particular, of supplying missing premises. (This is one of the reasons

that RST is often an inappropriate tool for analysing argumentation, (Reed and Long, 1998;

Reed, to appear)). In argument analysis, however, argument reconstruction forms a critically

important phase (van Eemeren et al., 1993), because arguments are typically heavily

abridged. Enthymemes – arguments (or, specifically, syllogisms) with one or more

components left implicit – are extremely common, to the extent that the natural language

expression of a Modus Ponens argument (A, A implie s B, therefore B) is so frequently

contracted through the omission of the major premise (leaving just: A, so B) that it has led

linguists to regard it as a separate form of argument altogether – the Modus Brevis (Sadock,

1977). Though there are various patterns to these contractions (such as those described by

Sadock and in (Reed, 1999b)) the software itself needs only to support an analyst’s work at

reconstruction. This support is provided in allowing new premises not explicitly included in

the text to be added to the structure of an argument.

The emphasis upon comparison with Rhetorical Structure Theory is quite deliberate. By

accepting the diversity not only of language, but also of the interpretation and analysis of

language, RST has become a powerful and widely used tool in discourse analysis and

computational linguistics, and has played a key role in making common resources available to


7

the research community. By equipping argument analysis tools with a similar flexibility and

tolerance of analytic diversity, the rich variety of approaches in teaching, learning, and

research can be preserved whilst at once providing a common interlingua and environment for

carrying out those activities.

As part of the commitment to supporting diversity, Araucaria has been developed in Java, to

support execution on many platforms. The software has been tested under various versions of

Microsoft Windows, Solaris, Linux and MacOS.

System Overview

The system can load either a text file or an existing, marked up, argument. In either case, the

left-hand pane shows the original text of an argument. Selecting part of this text and dropping

the selection on to the right-hand, diagram, pane creates a node which corresponds to that

text. There are then a number of actions which structure and combine the nodes in the

diagram:

ïDragging a line from one node to another creates a support relationship, indicated by an

arrow from premise to conclusion (supporter to supported)

ïMultiple lines of support can be selected and 'linked' together (and then subsequently

unlinked back to convergent support – Araucaria adopts the terminology of Freeman

(1991), inter alia)

ïSelecting either multiple lines of support or multiple nodes allows the selection of an

argumentation scheme to be associated with those nodes. Schemes are identified by

coloured areas on the graph, and can be overlapping (that is, a single text span can play a

role in several argumentation schemes – typically as a premise in one and a conclusion in

another)

In addition, the menus also support a range of further functionality:

ïThe addition of missing premises to the diagram

ïThe creation, modification, loading and saving of argumentation scheme sets

ïThe deletion of components from the diagram

File manipulation supports the saving not only of the marked up argument using XML, but

also of the diagram itself as a JPEG image for incorporation into documents and online

material.

Finally, as opinion seems divided in both the research and pedagogic communities, the entire


8

diagram can be inverted at any time with a single key-press.

Figure 1. Araucaria

4.AML

The environment provided by Araucaria is one suited to analysis. That is, it is assumed that a

sample text is available and that this text is to be analysed to produce a diagram. It is

important during this process that links between the original text and the corresponding

components of the diagram be maintained: deletions of components of the diagram, for

example, rely upon these links. Further, once the analysis is complete, it is useful to save not

just the text and diagram, but also the relationship between them, to allow future modification

and manipulation.

These factors suggest that marking up the original text is an appropriate approach. By adding

to the original text tags that indicate the evolving structure, the relationship between text and


9

diagram is preserved. The argument markup language (AML) defines a set of tags that

indicate delimitation of argument components (loosely, propositions), tags that indicate

support relationships between those components, and tags that indicate the extent of instances

of argumentation schemes.

Both Araucaria, and the markup language in which analyses are saved, exploit the typical tree

structure of argumentation1. This means that the markup language can be defined quite

succinctly, by characterising an argument recursively as a proposition supported by one or

more arguments. Implementation of the markup language employs XML (eXtensible Markup

Language) which carries with it a range of benefits. Firstly, XML is a well-defined, well-

understood, widely used industrial standard. This means that there are a range of standard

tools which can be employed to manipulate the data: one simple example is a sample

application available with many XML parsing suites2. The application provides a

conventional tree view (such as is employed by Microsoft Windows Explorer) that, in the

context of the files produced by Araucaria, allows a conclusion to be expanded to show its

(immediate) supporting premises, and each of those premises to be expanded to show their

supports, and so on. There are similarly tools for creating summaries, diagrams, for verifying

content, etc.

Secondly, as a generic data representation language XML files are also easily translatable into

other formats through the application of stylesheets. A good example of the possible uses of

1Argument analysis diagrams are not always trees (i.e. a node is not always restricted to having a

single parent). Setting aside circular reasoning, which can be handled as an argumentation scheme, non-tree structures are required where a single premise supports multiple conclusions – Freeman's (1991) divergent structures. These structures are unusual from a formal perspective (see (Shoesmith and Smiley, 1978) for a formal investigation), and are also rare in argumentation practice, with many textbook authors ignoring them altogether (see, e.g., (Groarke et al., 1997)). Given the overwhelming advantages of adopting tree rather than graph structures, this is an acknowledged restriction of the Araucaria system. A straightforward workaround in the situation where premise P1 supports conclusions C1 and C2, is to replicate P1, once in its role of supporting C1, and once in its role of supporting C2. There is, in general, no restriction on the uniqueness of premises. The underlying markup language is also restricted in respect of its handling of divergent argumentation, again for pragmatic reasons (parsing the data is quicker and more efficient if a tree structure can be exploited), though it is less of a problem here. Application software – such as Araucaria – that reads or writes to the format could be designed in such a way that divergent structures are created and displayed in the conventional manner, with P1 supporting C1 and C2, but then saved with the replication mentioned above. The development of application software which can handle divergent structures in this way is left to future work.

2The suite referred to here is that provided by Apache at http://xml.apache.org including, amongst other components , the Xerces parser. The application mentioned in the text is called “Treewalker” in the Xerces software.


10

stylesheets is in the creation of tailored HTML web files. Thus arguments can be

automatically summarised or made navigable for online provision. The discussion returns to

various such applications in sections 5 and 6, below.

Thirdly, XML has recently been recognised as having the potential to play a significant role in

corpus resources (Ide, 2000). Since this is one of the areas of application planned for

Araucaria and AML as described in section 6, the adoption of XML leads to several

advantages in the development of, publication of, and access to, corpus resources.

Fourthly, the acceptance of XML as a de facto industry standard facilitates data sharing: with

a single common format or interlingua, separate applications can share data in the tasks of

input, manipulation and output. With the structure of a particular XML markup language

defined independently in a document type definition (DTD) file, the definition of AML can

quickly and simply be made open to the community through the publication of its DTD .

Finally, the definition of AML in its DTD is independent of the applications in which it is

employed. This independence facilitates maintenance and development, whilst permitting

backward compatibility. AML is defined in argument.dtd, a full specification of the

components from which arguments – and argument diagrams – can be constructed. (The file

argument.dtd can be downloaded from the project homepage at

http://www.computing.dundee.ac.uk/staff/creed/research/araucaria).

Work is under way to extend the DTD in several specific directions, as described in section 7,

but these extensions have been planned during the design of the current version of

argument.dtd. Arguments marked up according to later formats will thus be compatible

with earlier data definitions: that is, newer formats will only add to and extend – and not

otherwise modify – earlier formats.

The Araucaria software is thus uncoupled from the AML; the latter can evolve

(monotonically) without requiring changes to the former3. Furthermore, the stability of the

AML also facilitates the development of a suite of related applications clustered around the

AML hub.

3The current released version of Araucaria is still loosely coupled to the AML definition as it carries

out AML parsing statically. The code is currently being migrated to exploit a SAX parsing module which dynamically tailors the parsing process to the specified DTD.


11

5.The Broader Context

The argument markup language AML has been designed to form a flexible data

representation scheme that can be employed in a number of applications. In this section, the

roles and relationships between these applications is sketched.

Figure 2. AML at the hub of a suite of modular applications

AML forms the hub, acting as a common data representation format that each application can

write to, or read from: applications on the left are involved in the creation of AML content;

those on the right in the manipulation and display of existing AML material.

Creating AML content

Currently, the primary means of creating AML content is Araucaria: that is, providing a

simple application to support the manual markup of text. Though perhaps the most obvious

approach, a range of alternative methods of supporting content creation are also under

investigation.

One ambitious means of creating AML material is to automate the process of marking up.

Work on this task is currently at an early stage, primarily because the broader research

community has only recently started to investigate systematically the relationship between a

variety of cues (such as discourse markers and punctuation) and argument structure (Snoeck

Henkemanns, 1997). The complexity of the task is also compounded by the problems posed

AML

Araucaria

Dialogue

Semi-automation

Araucaria

Dialogue

OnlineHTML


12

by enthymemes: automatically reconstructing implicit premises is a task which skilled

analysts find both taxing and hard to explain. Automation of this task thus represents a

significant challenge. A more reasonable short-term goal is to link the human analyst and

computer tools, such that the mistakes of one might be flagged by the processing of the other,

with their respective skills working in a complementary fashion.

A second project is investigating the role of dialogue in creating argumentation structures. In

building applications which rely upon domain knowledge (expert systems, decision support

systems, etc.), one of the most challenging tasks is knowledge elicitation: acquiring

appropriate information from experts. Recent work has suggested that dialogue logics can be

employed as a means of structuring the problem. Freeman (1991) proposes a dialogic

interpretation of argumentation structures that is essentially a simple, stylised dialogue logic.

He suggests that it is possible to view the move in a monologue from claim to data, or from

conclusion to premise, by conceiving of an imaginary interlocutor who asks the question,

'Why?' - the ground adequacy question. Moving from one premise to the next in a convergent

arrangement (i.e. with independent support) is the result of the implicit question, 'Can you

give me another reason?', and moving from one premise to the next in a linked arrangement,

the result of 'Why is that relevant?'. The 'dialogue logic' that results is thus rather

impoverished by comparison to, say, Mackenzie's (1979) DC or Walton and Krabbe's (1995)

PPD and RPD, but nonetheless it shares the core features of sets of rules specifying what can

be said when, and what the interlocutors are rationally committed to. Of course, like RPD

inter alia , the game is very much one-sided, with one player naively probing with one of three

questions, and the other player providing all the information. The prototype system developed

implements the dialogue logic by having the machine take on the role of question-asking

player, leaving the user to respond appropriately. Thus to any claim that the user enters, the

machine will probe for supporting premises and claims of relevance for those premises; if the

user has no premise available she can enter a blank and the machine will halt probing in that

particular direction. The following excerpt (Ex2) demonstrates a typical exchange:

(Ex2) User (A): Course ac2201 is interesting

System: Why is it that course ac2201 is interesting?

User (B): ac2201 involves studying computer hardware

System: Why is "ac2201 involves studying computer hardware" relevant to

"Course ac2201 is interesting"?


13

User (C): Looking at hardware means taking a hammer to a computer

System: Why is it that looking at hardware means taking a hammer to a

computer?

User: -

System: Why is it that ac2201 involves studying computer hardware?

User (D): Students leaving second year must have a thorough understanding of

the fundamentals of computing

...

System: Can you give me another reason for why course ac2201 is

interesting?

User (E): ac2201 teaches students how to use Linux

System ...

By carrying out the exchange in (Ex2), the system builds up an argument diagram (that is, an

AML description of an argument) shown in Figure 3:

Figure 3. Argument diagram for ac2201 argument

Many issues of discourse coherence, structure navigation, and linguistic manipulation remain


14

to be addressed, but the prototype clearly demonstrates that an argumentation structure can be

built up through a reasonably natural dialogue. The key benefit of the AML approach is that

the exercise is not simply an end in itself. The argument created by this dialogue (perhaps an

exchange with a member of faculty) could then be employed by a separate application to offer

advice to students as to which courses to sign up for.

Manipulating AML content

Currently the primary means of viewing and manipulating AML material is once again

Araucaria. There are, however, two distinct ways in which Araucaria can be employed in this

regard. The first is as described above, with an interactive application which loads an AML

file and allows the user to view and modify the argumentation structure. In addition,

Araucaria can be employed in a stripped down version simply to load and view an argument

diagram. More than just enforcing 'read-only' access, this method is useful in providing an

online applet: work is underway to provide a database of AML files which can be accessed

over the WWW, and displayed using this cut down applet version of Araucaria. This forms an

implemented platform upon which to develop corpus resources, as described in the next

section.

Again, there are also several other applications providing alternative means of access to AML

content. The first exploits the capabilities of stylesheets in dynamically producing HTML

content from AML data. Work is under way to produce a library of such stylesheets,

supporting a range of different views of arguments. Some of these views act as summaries

(showing only part of a whole argument), and others as typographical improvements

(indenting and formatting to highlight various structural aspects). Yet others in this library

perform more sophisticated transformations, producing versions of source arguments in which

individual premises and conclusions appear as hyperlinks, from which supporting data can be

linked, thus providing a means of navigating a large or complex argument. In this way a

single component of this stylesheet library is currently capable of producing HTML output

which attempts to mimic the format (though certainly not the content) of Scaltsas' extensive

online Archelogos project (Scaltsas, 1997).

Finally, dialogue can also be exploited as a means of content provision, albeit in a restricted

form. In much the same way as data can be acquired by the machine through implementation


15

of a dialogue logic with the machine as the questioner, so too can data be provided by having

the user take on the role of questioner. Clearly, to conform to a specific dialogue logic, it is

necessary to restrict the user's input, so that the bounds on questioning are clearly identified.

Prototype work in implementing Walton and Krabbe's RPD has demonstrated that supporting

user participation in tightly constrained dialogue logics is both feasible and an acceptable

model for interaction, and work is now under way to provide a user with the ability to probe

an existing AML structure with simple questions following the pattern of Ex2.

6.Application Areas

There is a very wide range of potential applications of the various components described in

the previous section; in this section, the focus is squarely upon the current and potential

applications of the currently implemented system functionality.

The foremost application domain is pedagogy. Teaching critical thinking skills, particularly in

North America, forms an important part of the curriculum in providing generic, transferable

skills. Syllabi for the topic, such as those provided in popular text books such as (Johnson &

Blair, 1993; Govier, 1997; Groarke et al., 1997) typically introduce some method for

diagramming arguments fairly early on, to provide students with the practical scaffolding

around which to erect a battery of analytic techniques. Though the various techniques may

differ somewhat, and the presentation of them differ significantly in these works, the

diagramming tools are substantially the same. Thus the diagramming itself is uncoupled from

the subsequent presentation of critical thinking skills and techniques, which suggests that a

‘theory-neutral’ software tool such as Araucaria might be successfully employed as a

component of teaching support in a broad range of argumentation and critical thinking

courses. The current version of the software is being trialled on undergraduate courses at

several Canadian and US universities in the fall of 2001.

One problem that Araucaria does not yet address is that of computer based assessment. In

some situations, the burden of marking on academic staff can be drastically reduced through

the use of automated marking software. Typically, such software relies upon a very narrowly

bounded range of possible answers, and although some of the problems associated with

traditional “multiple choice” papers can be avoided (Scriven, 2001), such assessment

techniques nevertheless severely restrict the freedom of the student. The approach offered by


16

Araucaria, in contrast, offers the possibility for automatic marking of students’ unfettered

argument analysis by exploiting techniques for graph matching (Bunke & Messmer, 1997).

This avenue is left to future work, for it involves one key challenge: that the comparison

between student and model answers should be sufficiently flexible to handle the wide

variation in argument structures which might be considered ‘correct’ or ‘nearly correct’.

With domain information structured as arguments in AML, there is also a rich potential for

supporting the teaching of other topics. Thus, for example, a small corpus of AML arguments,

perhaps constructed using Araucaria, could capture the material for part of an introductory

paleontology course, giving arguments for and against conclusions to be drawn from various

aspects of the fossil record4. Student interaction with this resource could then form part of a

computer assisted learning environment such as those described in (Jackson, 1998).

Employing applications such as those described in section 5, students could review arguments

and summaries of them, engage in dialogic exchanges, extend the existing arguments with

their own addit ions, and so on.

In a similar vein, such applications could also have a role to play outside the classroom, in

providing one resource for topics in the Public Understanding of Science and Technology

(PUST). With online provision of the same tools (Araucaria, dialogic interaction, dynamic

generation of summaries, etc.) and a set of AML resources in topical areas such as the genetic

modification of food, conflicting viewpoints could be presented to the public in a coherent

and measured way. It would be simple to provide for public interaction with the arguments,

supporting the contribution of new arguments to the online database. Using arguments to

structure online debate has been found to be a good means of involving people in public

policy decision making processes (Gordon & Karacapilidis, 1997), and it might be expected

that similar advantages might accrue in PUST.

Araucaria and the underlying data representation format also has the potential to serve the

academic community in several respects. First, from a practical point of view, the output of

Araucaria, both graphical and textual, simplifies the task of preparing material for

dissemination. Secondly, and equally practically, having a common format in which to

4Paleontology as a discipline offers particularly rich examples of texts in which dialectical structure

and chains of argumentation are extremely clear. It is for this reason that many introductory texts employ a chronological basis for exposition, following the various turns of the academic dialogue. See, e.g. (Edey & Johanson, 1990) and chapters 15-16 of (Wilford, 1985)


17

express the structure of an argument simplif ies the task of exchanging and dissecting

analyses. (Argthry, 2001) has many good examples of academic discussion on structural

analyses of problematic examples; each suggestion and counter-suggestion is phrased in

idiosyncratic and lengthy analyses which increase the chances of misunderstanding and error.

A common language may not make the analyses themselves any easier, but it will at least

make the subsequent exchange and discussion of those analyses more open, and less prone to

confusion.

Much more substantial than these, however, is the provision of a corpus of argumentation,

analysed and marked up in AML, and made available online. Work is starting at Dundee to

construct such a corpus, the contents of which will be accessible from the WWW, with

Aracuaria as an applet for viewing individual arguments. Because the data is stored in a

highly structured form using AML, sophisticated access and manipulation becomes possible:

a visitor might search for arguments with a particular structure, or for examples of a particular

argumentation scheme, or for arguments in a particular domain, or for arguments with a

particular degree of complexity, and so on. Furthermore, if Araucaria is used by both

academics and students in argumentation to mark up new examples of arguments, then these

analyses can be submitted to the corpus to extend the resource for others.

7.Conclusions

The Araucaria system performs a range of functions which are unique, and the software has

the potential to play a significant role in both academic and educational domains. Perhaps the

most similar software is van Gelder’s (2001) Reason!Able system. Like Reason!Able,

Araucaria employs a tree structure for mapping out the relationships between components in

an argument, and allows the user to manipulate that structure. Unlike Reason!Able, however,

Araucaria is driven primarily by research concerns rather than educational concerns, and

although pedagogy is a significant application area for Araucaria, it is not the only such area.

As a result, the current version utilises, for example, recent research in argumentation

schemes to provide support for analyses based upon such schemes. In addition, Araucaria also

starts with the assumption that the task at hand is one of analysis of existing argument, rather

than the construction of a new argument; for Reason!Able the focus is squarely upon

argument synthesis. Finally, Araucaria differs fundamentally from Reason!Able and all other

argumentation software in its provision of AML, an open standard for argument description


18

defined in XML, which has the potential to have a pervasive effect in both teaching and

research.

One feature that is offered by Reason!Able but is not currently available in Araucaria is the

notion of rebuttal, namely, that a particular premise might function in detracting from, rather

than supporting, a conclusion5. Future work aims to address this issue in a principled way in

two steps: first by extending AML, and then by developing a version of Araucaria which

exploits the AML extensions. The fundamental restriction that is currently imposed on

Araucaria is that the design of AML assumes that an argument is monologic, such that no

alternative or dissensual views can be represented. The first step, then, is to extend AML (in

such a way that it is backwardly compatible with the current version, as described in section 4

above), so that it can represent multiple points of view, thereby characterising dialogic

argument. The second step is to determine how best dialogic argument can be represented

graphically by a tool such as Araucaria. Both tasks are currently under investigation.

Though many of the applications described in section 6 are currently under development, they

are described here to show the role that the implemented Araucaria system and its underlying

data representation format, AML, will play in a variety of domains. Though Araucaria on its

own represents a significant tool for those working in argumentation, when coupled with the

applications in the domains suggested, it has the potential not only to play a key role in the

development of a range of systems of real utility in academic, pedagogical, and public arenas,

but also to support and encourage the further development of aspects of argumentation theory

and the application of that theory in computer systems.

Araucaria is free, open-source software, released under the GNU General Public License. The

software can be downloaded from the project homepage,

http://www.computing.dundee.ac.uk/staff/creed/araucaria/

Alternatively, a distribution CD can be obtained by sending an email to

[email protected]

5In fact, it would be possible to characterise rebuttal as a particular argumentation scheme, but this is a

rather unprincipled solution which would be very cumbersome to extend to truly dialogic


19

Acknowledgements

The authors would like to clearly acknowledge the work of three of Chris Reed's students,

Simon Wells, Yang Li and Kaifeng Li. Simon worked on a fina l year undergraduate project in

2000/01 focused upon the implementation of RPD, and Yang and Kaifeng worked during

2001 on M.Sc. projects implementing the prototype application for dialogic creation of AML

material, and a number of XSL stylesheets for the dynamic generation of HTML content from

AML, respectively. It is their work to which reference is made in section 5.

Grateful acknowledgement is also made of comments and feedback on this work offered by

Leo Groarke, David Hitchcock, and Doug Walton. Any errors that may remain, however, are

of course entirely the responsibility of the authors.

References

Argthry (2001) Discussions on the ARGTHRY list, particularly the archives for July 2001

(available through the process described at http://www.yorku.ca/gilbert/argthry/)

Beardsley (1950) Practical Logic. Prentice Hall.

Bunke, H & Messmer, BT (1997) "Recent advances in graph matching", International

Journal of Pattern Recognition and Artificial Intelligence, 11 (1): 169-203.

Edey, M.A. & D.C. Johanson (1990) Blueprints: Solving the Mystery of Evolution, Oxford

University Press.

Freeman, J.B. (1991) Dialectics and the MarcoStructure of Arguments, Foris.

Gordon, T.F. & Karacapilidis, N. (1997) “The Zeno Argumentation Framework” in

Proceedings of the 6th International Conference on AI and Law (ICAIL’97).

Govier, T. (1997) A Practical Study of Argument, Wadsworth.

Grennan, W. (1996) Informal Logic: Issues and Techniques, McGill-Queens University Press.

Groarke, L., C.W. Tindale & L. Fisher (1997) Good Reasoning Matters, 2nd Edition, Oxford

University Press.

Ide, N. (2000) “The XML Framework and Its Implications for Corpus Access and Use” in

arguments.


20

Proceedings of LREC’2000 MMLR Workshop, Athens.

Jackson, S. (1998) “Beyond Web Course Automation: Designing Online Dialogue” in

Working Notes of NAU/Web’98

Johnson, R. & J.A. Blair (1993) Logical Self Defense, McGraw Hill.

Kienpointner, M. (1992) “How to classify arguments” in F.H.van Eemeren, R. Grootendorst,

J.A. Blair & C.A. Willard, Argumentation Illuminated.

LeBlanc, J. (1998) Thinking Clearly, W.W.Norton & Co.

Mackenzie, J.D. (1979) “Question Begging in Non-Cumulative Systems” Journal of

Philosophical Logic 8:117-133.

Mann, W.C. & S.A. Thompson (1988) “Rhetorical Structure Theory: A Theory of Text

Organisation” Text 8 (3) 243-281.

O’Donnell, M. (1997) “RST-Tool: AN RST Analysis Tool”, in W. Hoeppner (ed.)

Proceedings of the 6th European Workshop on Natural Language Generation

(EWNLG’97), Duisburg.

Perelman, Ch. and L. Olbrechts-Tyteca (1969) The New Rhetoric, Notre Dame Press.

Reed, C.A. (1997) “Representing and Applying Knowledge for Argumentation in a Social

Context” AI & Society 11:138-154.

Reed, C.A. (1999) “Building Monologue” in Proceedings of the 3rd OSSA Conference

(OSSA’99), St. Catharines.

Reed, C.A. (1999b) “The Role of Saliency in Generating Natural Language Arguments”, in

Proceedings of the 16th International Joint Conference on Artificial Intelligence

(IJCAI’99), Morgan Kaufmann, pp876-881.

Reed, C.A. (to appear) “Reconciling argumentation theoretic and discourse relation based

approaches to natural language argument”

Reed, C.A., Norman, T.J. & Gabbay, D. (2002, to appear) Argument and Computation

Reed, C.A. & D.P. Long (1998) “Generating the Structure of Argument” in Proceedings of

the 36th ACL and 17th COLING (ACL-COLING’98), pp1091-1097.

Reed, C.A. & Walton, D.N. (2001) “Applications of Argumentation Schemes” in Proceedings

of the 4th OSSA Conference (OSSA’2001), Windsor.


21

Rips, L.J. (1994) The Psychology of Proof, MIT Press

Sadock (1977) “Modus Brevis: The Truncated Argument” in Papers from the 13th Regional

Meeting of the Chicago Linguistics Society pp545-554

Scaltsas, T. (1997) “Representation of Philosophical Argument” in T. Bynum and J. Moor

(eds) The Digital Phoenix: How Computers are Changing Philosophy, Blackwell.

Scriven, M. (2001) “Testing Critical Thinking” in Proceedings of the 4th OSSA Conference

(OSSA’2001), Windsor.

Shoesmith, D.J. & T.J. Smiley (1978) Multiple Conclusion Logic, Cambridge University

Press.

Snoeck Henkemanns (1997) “Verbal Indicators of Argument and Explanation” in

Proceedings of the 2nd OSSA Conference, St. Catharines.

Thomas, S.N. (1986) Practical Reasoning in Natural Language, 3rd Edition, Prentice Hall.

Toulmin, S.E. (1958) The Uses of Argument, Cambridge University Press.

van Eemeren, F.H., R. Grootendorst & F. Snoeck Henkemanns (1993) Reconstructing

Argumentative Discourse, LEA.

van Gelder & A. Rizzo (2001) “Reason!Able Across the Curriculum” in Is IT an Odyssey in

Learning? Proceedings of the 2001 Conference of the Computing in Education Group

of Victoria.

Walton, D.N. (1996) Argumentation Schemes for Presumptive Reasoning, LEA.

Walton, D.N. & E.C.W. Krabbe (1995) Commitment in Dialogue, SUNY Press.

Wilford, J.N. (1985) The Riddle of the Dinosaur, Faber and Faber.

Wilson (1986) The Anatomy of Argument, University Press of America.

Documents

Araucaria: Software for Puzzles in Argument Diagramming